1. Field of the Invention
The present invention generally relates to the field of semiconductor manufacturing, and more specifically to crystallographic texture measurement and analysis systems for polycrystalline materials on wafers.
2. Background of the Invention
The physical properties of single crystals, such as electric, elastic, and magnetic properties, are directionally dependent and usually represented by tensors of the second, fourth and sixth order, respectively. As a consequence, a polycrystalline material, which is an aggregate of single crystals (called grains or crystallites), has anisotropic properties. The degree of anisotropy of a macroscopic specimen depends on the orientation distribution of its crystallites, or texture, with respect to the sample fixed coordinate system.
As an example, most thin film metallization processes for semiconductor applications result in a preferred orientation of grains with respect to growth surface. The crystallographic texture of thin films and discrete structures used in integrated circuits greatly affects their reliability and performance, and may be controlled by tunable manufacturing processes. A discussion of the importance of texture and disclosure of texture control methods is found in the paper titled "Microstructure Control in Semiconductor Metallization", J. M. E. Harper, K. P. Rodbell, J. Vac. Sci. Technology, 15 (4), 763-779, (1997).
The quantitative measure of texture can be described by the so-called Orientation Distribution Function (ODF) which permits one to describe texture in a rigorous mathematical way and to calculate the macroscopic properties from the corresponding single crystal properties. However, a direct method of ODF measurement has not been developed. The experimental determination of ODF is currently only performed by destructive and time-consuming measurements of orientation and volume of large numbers of individual grains, subsequent mathematical analysis of which then yields a unique ODF for the sample of interest.
X-ray diffraction is a well-known technique for measuring the physical properties of polycrystalline materials. See, for example, H. P. Klug, L. E. Alexander, "X-ray Diffraction Procedures", Wiley & Sons, (1974). X-rays diffracted from the surface of a polycrystalline material provide direct information about the size, spacing, and orientation of crystallites that comprise the polycrystalline material. X-rays impinging on the material will scatter in all directions. Constructive interference of the scattering x-rays occurs only at particular angles that the scattering x-rays make with the incident x-ray beam, and is dependent on the crystalline spacing and orientation. This information is represented in the form of diffracted x-ray intensity versus the diffraction angle from incident beam. Constructive interference of scattered x-rays from the crystalline structure results in intensity maxima, also referred to as diffraction peaks. Each particular set of crystalline structures of a material will have an associated diffraction peak that occurs at a particular angle.
There exist numerous commercial x-ray diffraction instruments that measure the physical properties from which the texture of polycrystalline materials may be determined, including those produced by Philips Analytical X-ray, Bruker AXS, Rigaku International Corp., Scintag Inc., Bede Scientific Inc. and others. However, use of these systems for texture determination, while feasible, is nonetheless time-consuming, lacks sufficient resolution, and is limited to relatively small semiconductor wafers. Furthermore, these current systems contain inherent limitations that make their conversion to a rapid, high precision measurement tool for large uncut wafers (e.g., 200 millimeter diameter, and recently introduced 300 millimeter diameter) problematic. The speed of measurement, obtainable measurement resolution, and applicable wafer size remain as limitations of the state of the art in texture analysis of semiconductor wafers.
Using an area x-ray detector on an x-ray diffraction instrument increases the speed of texture analysis considerably, but area x-ray detectors are not currently used as efficiently as possible for texture analysis. This is primarily due to the fact that they employ traditional texture analysis protocols that do not efficiently use all of the diffraction information captured by the area detector. The resultant measurement time is still quite long. As a consequence, fewer samples are typically analyzed due to the excessive measurement times required.
Motion control systems have been built into x-ray diffraction systems for the mapping of texture over the surface of a large (e.g., 150 millimeter) wafer, but they too are not designed to make the most efficient use of diffraction information obtained from the area detector. Such motion control systems also tend to be complex and very expensive. Even if current x-ray diffraction systems utilizing area detectors were converted to map texture in larger wafers (by increasing the size of the texture mapping stages), they would be extremely inefficient, complex, costly, and slow. Efficient integration of a wafer motion system (for mapping texture over the entire wafer surface) with an area x-ray detector has not been achieved with the currently available instruments.
The current texture analysis methodologies additionally are not suitable for new generations of materials (such as polycrystalline and epitaxial films, or superconductors) that have sharp textures. The methods lack the required resolution, do not take advantage of the sample and crystal symmetry in order to expedite testing, and do not take advantage of modern computing capabilities. The current texture analysis methodologies do not make efficient use of all the diffraction data captured on an area x-ray detector. While texture analysis using current commercial x-ray systems is versatile, it is tedious, slow and requires a highly trained operator. The present systems are not practical for large sample throughput rates and automated operation, as would be required for commercial manufacturing operations.
Considering texture analysis and the state of the art in more specific detail, the Orientation Distribution Function (ODF), the quantitative measure of texture, can be described in G-space, where the orientation g of an individual crystallite with respect to the reference system of the sample is described by three independent parameters (usually angles) g={.alpha., .beta., .gamma.}. A schematic representation of G-space is shown in FIG. 1. A direct method of ODF measurement does not exist. The experimental determination of ODF, as mentioned hereinabove, is possible through destructive and time consuming measurements of orientation and volume of large numbers of individual grains, which yield a unique ODF (see K. Lucke, H. Perlwitz, and W. Pitsch, "Elekronenmikroskopische Bestimmung der Orientierungsferteilung der Kristallite in gevaltztem Kupfer," Phys. Stat. Sol. 7 (1964), 733-746, and F. Wagner "Texture Determination by Individual Orientation Measurement," in Experimental Techniques in Texture Analysis, ed. H. J. Bunge (DGM, Oberusel, 1986), 115-124).
The more practical and nondestructive experimental method is the direct measurement of a volume of crystallites with two of the three angular parameters fixed and the third parameter varied through all possible values. This is the so-called pole figure measurement. The ODF is subsequently calculated from several pole figures (pole figure inversion). The term "pole figure" is understood as the intensity distribution of a certain physical quantity in reference to the sample coordinate system. The pole figure measurement is most commonly obtained by diffraction (of x-rays, neutrons or electrons). The measured physical quantity in this case is the intensity of x-rays diffracted from a particular set of crystallographic planes. The central problem of quantitative texture analysis is the reproduction of the ODF from experimental pole figures. The original works of Bunge (H. J. Bunge, Texture Analysis in Materials Science (Butterworths, London, 1982)), Roe (R. J. Roe, "Description of crystalline orientation in polycrystalline materials. (III) General solution to pole figure inversion," J. Appl. Phys. 36 (1965), 2024-2031) and Matthies (S. Matthies, "On the reproducibility of the orientation distribution function of texture samples from pole figures (ghost phenomena), Phys. Stat. Sol. (b) 92 (1979), K135-K138) have been followed by newer methods (K. Pawlik, Phys. Stat. Sol. (b) 124 (1986), 477).
The experimental methods of pole figure measurement are classified into two groups: 1) a constant diffraction vector method and 2) a variable diffraction vector method. In the constant diffraction vector method, the sample is rotated in such a way that each sample direction is brought successively into the direction of the diffraction vector, as shown in FIG. 2. This requires at least two independent sample rotations. Usually, the fixed diffraction vector method of pole figure measurement is performed on a four axis goniometer that allows for three independent sample rotations .omega., .phi., and .chi., and one detector rotation 2.theta. as shown in FIG. 3. The three rotations can be combined in many different ways in order to scan all the sample directions. The most common methodology is to fix .omega. and rotate .phi. in the range 0-360.degree. and .chi. in the range 0-90.degree.. In practice the maximum .chi. is reduced to 75-85.degree. due to high defocusing errors at higher tilt angles. Reducing the .chi. rotation to less than 90.degree. results in a so-called incomplete pole figure. A complete pole figure may be constructed by combining the transmission and reflection measurements as recommended in American Society for Testing and Materials Standard E81-96, "Standard Test Method for Preparing Quantitative Pole Figures". Such a method required that a very thin isolated foil of material being tested be prepared and therefor is not practical as part of a commercial test method. Only one crystallographic reflection is used at a time in the fixed diffraction vector method. In this case the point detector is fixed at a given 2.theta. corresponding to a particular crystallographic plane.
In the variable diffraction vector method, the incident and diffracted beams are rotated with respect to a sample. The sample is rotated only through one angle. Such an arrangement is commonly used in electron diffraction in electron microscopes. Both aforementioned methods of pole figure measurement measure only one pole figure at a time.
Even though the pole figures are continuous functions, their measurement is carried out in a sequence of discrete steps. One major drawback of the apparatus and measurement protocol described above is that it is relatively slow due to the large number of sequential scanning steps required with a point detector. Usually the pole figures are scanned in 5.degree. steps in .phi. and .chi. in order to keep the total measurement time at a reasonable level. Even so, measurements can require up to several hours. For example, there would be 72 .phi. positions (0.degree. to 360.degree. in 5.degree. steps) and 17 .chi. positions (0.degree. to 85.degree. in 5.degree. steps) for a total of 1224 measurement points in a single pole figure. Assuming a typical collection time of 10 seconds per position, the total data collection time would be 3.4 hours per pole figure. For the ODF calculation for a cubic material one usually needs three pole figures, which would result in 10.2 hours for data collection. This total collection time does not account for the necessary corrections of the raw experimental data, and the computing time to calculate the ODF, which traditionally has also been a long procedure due to limitations on available computing power. Thus, the excessive measurement times required to obtain pole figures independently using a point detector represent a significant shortcoming of the prior art. A further significant disadvantage resulting from the long measurement times required by this approach is that the 5.degree. step scan can miss significant information in highly textured materials. In such materials texture can change quite dramatically within a range of 10.degree.. Thus, higher resolution measurements and analysis are required.
The conventional method of pole figure measurement uses a scanning point detector fixed at 2.theta. position and which registers the integral intensity of diffracted energy (optical, x-ray, neutron, electron, etc.). In the last decade, position sensitive detectors have seen some limited application to texture analysis (L. Wcislak, H. J. Bunge, Texture Analysis with a Position Sensitive Detector, (Cuvillier, Gottingen, 1996)). More recently two dimensional (2-D) area x-ray detectors (also referred to as position sensitive detectors or PSDs) have been used to measure texture much faster than traditional point detectors by measuring a range of angular directions simultaneously {[H. J. Bunge, H. Klein, "Determination of Qunatitative, High Resolution Pole Figures with the Area Detector", Z. Metallkd., 87 (6), P 465-475, (1996)], [K. L. Smith and R. B. Ortega, "Use of a Two-dimensional Position Sensitive Detector for Collecting Pole Figures," Advances in X-Ray Analysis, 36 (1993) 641-647], [U. Preckwinkel, K. Smith, B. He, B. Schey, B. Stritzker, "Texture Analysis in Thin Films Using an Area Detector", Presented at 47th Annual Denver X-ray Conference Aug. 3-7, Colorado Springs, Colo., (1998)]}. The active face of the 2-D area detector can be thought of as a tightly packed planar array of micro-detectors, all of which operate simultaneously. U.S. Pat. No. 5,828,784 issued to Kurtz, discloses a representative area detector.
Rather than forcing a sequential scan of the sample in two angular directions (.chi. and .phi.) to create an intensity "map" as required by point detectors, the area detector allows concurrent measurement over a large range of one direction, namely the .chi. direction, while located at one position, as shown in FIG. 4. The range of .phi. covered by the area detector is much less. The ranges of .chi. and .phi. depend on the measurement geometry, and in particular on the detector to measurement point distance. However, the area detector can typically observe more than one diffraction peak (crystallographic reflection) at a time. FIG. 4 shows the diffraction arcs corresponding to three crystallographic planes, captured on a single collection frame of the area x-ray detector. Thus the area detector can provide a range of .chi. and .phi. values (though the range in .chi. is much greater than the range in .phi.) for several crystallographic planes in one collection frame, which is equivalent to collecting several incomplete pole figures simultaneously. This results in much shorter data collection time compared to a point detector, that must sequentially scan an equivalent angular range.
Another primary advantage of using an area position sensitive detector is the registration of complete peak profiles and background profiles, instead of just integral intensities at only one point, as from a point detector. Thus the intensity can be integrated over the entire peak width and the background scattering can be subtracted out. This results in greatly improved intensity counting statistics and reduction of the measuring error.
By way of illustrative example and not limitation, one type of area x-ray detector suitable for texture analysis in polycrystalline materials is a multiwire gas proportional counter, e.g., the device manufactured by Bruker AXS (Madison, Wis., 608-276-3047) and commercially available under the product name HI-STAR. Bruker also produces a complete x-ray diffraction system based on the HI-STAR detector. It is sold under the product name GADDS (General Area Detector Diffraction System). The HI-STAR detector features extremely high sensitivity combined with a total detection area 11.5 centimeters in diameter. Compared to other types of x-ray detectors, the multiwire gas detector has slightly poorer angular resolution (presently around 150 .mu.m), but for texture analysis, particularly on thin polycrystalline films, sensitivity is generally more important than angular resolution. The large 2-D area of the HI-STAR detector allows for the simultaneous collection of diffraction intensity over a large range of .chi. angles and 2.theta. angles. The x-ray diffraction data shown in FIG. 4 was collected on thin copper film using a HI-STAR detector.
The exact range of 2.theta., .chi. and .phi. covered by the area detector depends on its distance from the sample. As the area detector is moved closer to the irradiated spot on the sample, the angular range of 2.theta., .chi., and .phi. covered by the area detector increases. As the area detector is moved farther away from the irradiated spot on the sample, the angular range of 2, .chi., and .phi. covered by the area detector decreases.
The present texture analysis protocols and software for the commercial GADDS x-ray diffraction system accommodate analysis of only one reflection (diffraction arc) at a time like a traditional diffractometer, essentially forcing the user to go back and reconfigure the system to analyze texture on a second or third reflection in separate steps. Thus, even though the detector might be collecting information on several diffraction peaks in any one frame, the data are not analyzed simultaneously.
The ODF reproduction from pole figures can only be solved numerically in practice (as opposed to a analytical mathematical solution). The analysis, historically developed for structural and geological materials, has been optimized for 5.degree. steps in experimental pole figures. The primary reasons for the 5.degree. steps were a practical limitation in data collection times and limitations on computing power.
The commercially available software supplied with texture goniometers (Bruker AXS, Phillips, Rigaku, Scintag, Seifert) or as a software package only (PopLA {Preferred Orientation Package-Los Alamos, U. F. Kocks, J. S. Kallend, H. R. Wenk, A. D. Rollet, and S. I. Wright, Los Alamos National Laboratory}, Beartex {The Berkeley Texture Package, University of California at Berkeley}, LaboTex {Labosoft, Krakow, Poland) use pole figures measured with a 5.degree. resolution for the ODF analysis. The computation time is several minutes since the algorithms do not make use of modem computing capabilities and are not optimized for speed. Decreasing the step to 1.degree. would increase the computation time by a factor of 125, making it impractical for many of today's highly textured engineering/electronic materials. Examples of such materials are semiconductor thin films, particularly metallization layers, which often have very sharp textures. The ODF analysis for such materials requires a higher resolving power while collecting pole figures. For the metallization blanket films and interconnects, the required resolution is 1 degree or less. However, a suitably fast measurement will always be preferred for a practical commercial instrument. The methodology of calculating the ODF from high resolution pole figure data collected with variable resolution has not been developed nor published to date.
Most traditional x-ray texture goniometers are not capable of handling large size samples, or mapping over the surfaces of large samples. Recently some manufacturers of commercial x-ray diffractometers such as Bruker, Philips, Scintag, Seifert, and Rigaku have offered mapping stages (x, y, z translations, and .phi. rotation) that are built onto an Eulerian cradle. These stages are primarily designed for thin wafers, as used in the electronics industry. The Eulerian cradle is used for the sample .chi. rotation. An example of such a system is the Bruker D8 ADVANCE with a 1/4 circle Eulerian cradle option, and a 150 millimeter x-y mapping stage built onto the Eulerian cradle (Bruker 1997 Catalog Order No. B88-E00001). In this apparatus the x and y stages are mounted on top of a z stage (vertical linear motion), which is in turn mounted on top of a .phi. rotating stage. The entire combination of x, y, z, and .phi. stages are then mounted onto the 1/4 circle (90.degree.) Eulerian cradle to provide the sample .chi. rotation.
This mapping stage was built to handle semiconductor wafer diameters to 150 millimeter (6 inches). Such a stage could clearly be enlarged with proper design to handle wafers up to 200 millimeters or 300 millimeters in diameter. Diffraction systems built around this combination of wafer motion stages could also be set up to utilize an area x-ray detector such as HI-STAR. For example, the D8 ADVANCE with 150 millimeter wafer mapping and HI-STAR detector is commercially available. However, such a system would still exhibit several disadvantages. First, the total wafer motion control would be quite expensive in order to maintain wafer alignment, due to the use of the large Eulerian cradle which provides the sample .chi. rotation, in addition to supporting the other motion stages (x, y, z, and .phi.). This is demonstrated by the present high cost of the current Bruker 1/4 circle Eulerian cradle option with the built-in 150 millimeter x-y mapping stage. Making it large enough to handle 200 mm wafers (8 inches), or 300 mm wafers (12 inches) would only exacerbate that cost.
A second major drawback is the large clearance distance required between the area detector and the contemplated combination of sample motion stages. The area detector must be located relatively far from the sample in order to allow for rotational and translational freedom of all the sample motion stages. If the detector were placed too close to the motion stages, the sample (or motion stage) would collide with the detector at the sample's far range of travel. A large distance between the detector and measurement location on the sample results in a much smaller range of .chi. and 2.theta. captured by the area detector at any one location. Thus more sequential angular locations are required to obtain texture information, and the measurement takes longer to carry out. In essence the system becomes more and more like a traditional four circle scanning diffractometer using a point detector, and less of the area detector advantage is capitalized upon. If the area detector is placed very close to the measurement point the angular range of 2.theta. and .chi. covered by the detector will be quite large, however the motion of the wafer will be severely restricted resulting in highly truncated pole figures.
It therefore is an object of the present invention to provide a crystalline texture measurement system that operates rapidly; measures texture with a high degree of accuracy; is capable of taking texture measurements over the entire surface of large (e.g., 200 millimeter and 300 millimeter) semiconductor wafers; is amenable to automation; and can be operated by persons without extensive specialized skill and training.
It is another object of the present invention to provide quantitative texture information on a defined material present on a wafer (e.g., the metal interconnect on a semiconductor wafer) much more rapidly than is possible with any texture mapping system in the prior art.
It is still another object of the invention to provide a system enabling sufficiently rapid texture measurement to allow automated use of texture analysis in development and production quality control in commercial semiconductor processing operations.
A further object of the present invention is to provide texture mapping information anywhere on the surface of a wafer under test, even for very large (e.g., 200 millimeter and 300 millimeter) wafers.
A further object of the present invention is to provide texture mapping information with a high degree of resolution, preferably in steps of one degree or less in the .phi. and .chi. directions.
A further object of the present invention is to provide a texture mapping system that can be operated in a production environment, and by persons without specialized skilled or training.
Other objects and advantages will be more fully apparent from the ensuing disclosure and appended claims.